Prediction Research Of Water Quality In Aquaculture Based On Machine Learning Method

Since ancient times,there has been a saying in the proverb "Cultivate fish first to raise water".The quality of the water directly determines the quality and quantity in aquaculture.Precise forecasting and warning of aquaculture water quality can strengthen water quality management and reduce the occurrence of water quality disasters.The dissolved oxygen content and ammonia nitrogen in water are the two most important parameters in the farming process.The dissolved oxygen content has achieved online measurement based on the Internet of Things technology.It is difficult to achieve accurate prediction of dissolved oxygen due to the nonlinearity,large time lag,and complex coupling relationships of the data acquired by the monitoring system.At the same time,due to the extremely low content of ammonia nitrogen in the culture water,it is very difficult to achieve accurate on-line measurement of ammonia nitrogen in this situation.How to accurately estimate the ammonia nitrogen based on the easy-to-measure parameters is another problem to be solved in this paper.Therefore,the dissertation takes dissolved oxygen and ammonia nitrogen as the research object,and uses data preprocessing technology and machine learning technology to study the theory and method of online prediction and early warning of aquaculture water quality.(1)Design of data collection scheme for pond aquaculture In order to solve the problem of easily losing and incomplete water quality data,a water quality data collection program based on online and offline data was designed.Based on the pond size,shape,and aquaculture species cultivated in the pond,the collection program designed multi-angle,multi-level,multi-sensor(same-parameter)correction collection points for pond layout.In the end,the research achieved 12 dissolved oxygen,water temperature,pH,and wind direction.The program can be online access to dissolved oxygen,pH,water temperature and other 11 parameters;ammonia nitrogen spectrophotometer Nash reagent test,multi-bit manual collection of ammonia nitrogen data.(2)Spatio-temporal variation of water quality factors in pond aquaculture and analysis of factors’relationship.On the basis of data acquisition,a comprehensive analysis was made of the spatial-temporal variation of water temperature,dissolved oxygen,pH,ammonia nitrogen,and other factors.The analysis results show that the water quality factors are layered in the vertical direction.With the increase in depth,the content is reduced;in the horizontal level of the same layer.And,the surrounding position of the pond is higher than in the middle position.All water quality parameters show periodic changes in time.Through factor analysis,it was determined that there was a strong correlation between aquaculture water quality parameters and meteorological parameters,which laid the foundation for the selection of model data.(3)Study on data pretreatment methods for pond aquaculture water quality A data denoising method based on multi-sensor data fusion was proposed to solve the problem of missing data,invalid data,and data redundancy for online measurement of multi-sensor(same parameter)water quality parameters.This method can simultaneously collect dissolved oxygen data from multiple sensors.Perform fusion output to improve dissolved oxygen data quality.A wavelet noise reduction method based on layered soft threshold based on a single sensor measurement scheme was proposed to solve the data noise problem for single parameter sensor.This method can quickly and effectively denoise single sensor data.The experimental results show that the pretreatment method can effectively reduce the noise of sample data.Taking ammonia as an example,the evaluation index is 8.55%,21.75%and 18.12%higher than the other three methods,and the BIAS is decreased by 18.58%.40.64%and 35.27%and RMS decreased by 54.54%,98.73%and 99.55%.Respectively,the proposed method provided a reliable data basis for the prediction model construction.(4)Study on the prediction method of dissolved oxygen in pond aquaculture.The dissolved oxygen content in aquaculture is non-linear changing and susceptible to multiple factors.The traditional prediction method is not suitable for small samples and high dimensions.In this study,an improved particle swarm optimization algorithm for least squares support vector machines is proposed.This method improves the iterative speed of particle swarm by setting the speed and position of the variable to perturb the particle swarm.The particle swarm algorithm is used to optimize the regularization parameters of the least squares support vector machine algorithm and the coverage width of the kernel function.The results showed that the IPSO-LSSVM model decreased by 26.47%,41.88%and 40.93%compared with the standard LSSVM model,respectively.The correlation coefficient R2 of the algorithm was 0.9623,which enabled rapid and accurate prediction of dissolved oxygen in aquaculture.The dissertation will use the dissolved oxygen data collected by multi-sensors to fuse the pre-processed data as the training model training model.The prediction model RBFNN-IPSO-LSSVM has a 23.05%lower variance than the IPSO-LSSVM,and the MAPE is reduced by 21.42%.(5)Ammonia-nitrogen soft measurement method for pond aquiculture.The ammonia nitrogen content is extremely low in aquaculture water quality.And the characteristics of the water quality of ammonia nitrogen also have non-linear and multi-factor effects.An ammonia-nitrogen based particle swarm optimization extreme learning machine based on empirical mode was proposed.The method first uses the empirical mode method to decompose the natural mode of ammonia nitrogen and obtain the hidden layer node number of the limit learning machine.Then,the particle swarm algorithm is used to select the weight and offset of the input layer of the extreme learning machine.On-line measurement of water quality and meteorological parameters enables the estimation of ammonia nitrogen content.The results show that the EMD-IPSO-ELM method has lower evaluation index MAE,MSE and MAPE by 11.57%,26.02%and 14.41%than the standard ELM.At the same time,the paper uses layered soft-threshold wavelet noise reduction to preprocess ammonia nitrogen data.The WA-EMD-IPSO-ELM model is better than the EMD-IPSO-ELM model obtained from the original data training.The evaluation index MAE,MSE and MAPE decreased by 26.27%,48.35%and 30.74%,which fully proved that data preprocessing can effectively improve the prediction accuracy.(6)Design and implementation of forecasting and warning system for water quality in pond aquaculture.In order to verify the accuracy and applicability of the dissolving oxygen content and ammonia nitrogen forecasting algorithm put forward in the paper,an aquaculture networking monitoring platform was integrated to design and implement aquaculture water quality forecasting and early warning system modules.The module follows the existing system’s Java MVC development framework and MySQL database,designing and implementing the algorithm.This module implements data pre-processing functions for data acquisition,data retrieval,and data curve rendering,as well as the ability to predict dissolved oxygen and ammonia nitrogen.The system was applied at a research base in Zhejiang and the results were good.Based on the above research,the paper achieved a comprehensive data collection through the design of aquaculture water quality and meteorological data collection scheme,which laid the data foundation for model construction.For the problems existing in single-sensor data collection,by exploring data preprocessing methods,a highly targeted pre-processing method was proposed.At the same time,based on the effective data,the short-term prediction of dissolved oxygen and the on-line soft-sensing ammonia nitrogen estimation model were constructed to achieve accurate prediction of dissolved oxygen content and ammonia nitrogen.